SM 5 BSZ - Pulse response and filter shape.
(Oct 1 2000)

Pulse response and filter shape.

The analog filters that are present in the chain between the antenna and the A/D converters will stretch any very short pulse that arrives at the antenna input. This means that all pulses for which the spectrum at the antenna input is much wider than the bandwidth of the analog filters will be shaped by the analog filters to one and the same shape for all pulses. This pulse shape is often ugly because the filters are designed for steep skirts, not nice pulse response. In order to make subsequent processing (noise blanking) easier. It is a good idea to insert a digital filter in the signal path to make the pulse response better (and the spectrum response flatter).

TS520

Fig.1. Shows the spectrum response of a TS520. A 50 ohm dummy load is connected at the antenna input. The spectrum is not flat enough to allow more than about 1.5kHz search width for weak signals unless the variation of the noise floor level is compensated somehow.


Fig 1. Frequency response of TS520. Two stations leak into the receiver which is set at 7MHz and has a dummy load for antenna. The spurs at 50, 100, 150 and 620Hz are frequency stable and essentially one pixel wide. The 40m cw signals drift mainly because of the poor frequency stability of the TS520.
The pulse response of TS520 is shown in figures 2 and 3. The pulses have a repetition rate of 100Hz. It is clearly seen that pulses with a repetition rate of 300Hz or will not be separated.

Fig 2. Pulse response of TS520. Short pulses with a repetition rate of 100Hz are fed to the antenna input. The upper track is in logaritmic scale and the lower is in linear volts scale.

Fig 3. Pulse response of TS520. Short pulses with a repetition rate of 100Hz are fed to the antenna input in the same way as in fig.2. The upper track is in logaritmic scale and the lower is in linear volts scale, expanded 16 times (24dB).

Direct conversion receiver with a Delta44 audio board.

The direct conversion receiver two mixers in quadrature. was tuned for 7MHz. An amplifier with a bandpass filter for 7MHz was inserted between the direct conversion receiver and a 50ohm dummy load. The gain was high enough to make the white noise from the dummy load the dominating signal. The rf filter was wide enough to have a small effect on the spectrum shape.

The anti-alias filter may provide a nice roll off towards the spectrum ends besides their function to reduce the alias signal. Note that some of the component values you find in the anti-alias filter schematic diagram have been changed. Small adjustments allow different compromises between bandwidth and alias suppression. With the component values in the schematic diagram the bandwidth is larger and the skirts steeper compared to what you can see in the graphs below. The complete Linux DSP receiver has a digital filter that corrects the frequency response so differences in the analog filters are completely absorbed by corresponding changes in the digital filters. Different choices for the anti alias filter function therefore affect dynamic range, not the final bandwidth.

Fig.4 shows the frequency response. Already without any correction the useful bandwidth is nearly 70kHz. If the anti-alias filter was omitted the spectrum would be completely flat but then the noise floor at the spectrum ends would contain contributions from both the desired frequency and the alias frequency. With the anti alias filter used to make these graphs the correcting digital filter will have up to about 15dB extra gain at the spectrum ends to extend the flat region to above 90kHz. The noise figure will be slightly degraded towards the ends but on the other hand the tolerance for strong undesired signals will be improved there.


Fig 4. Frequency response of direct conversion receiver tuned at 7MHz. This curve is essentially determined by the anti-alias filter.
Because of the flat center region and nicely curved end regions the pulse response of the direct conversion receiver is quite good as can be seen in figs 5 and 6. The A/D board is not DC connected so the DC component has to be zero always. The very long tail after the pulse is the time constant of the DC blocking of the A/D board.


Fig 5. Pulse response of direct conversion receiver tuned at 7MHz. Short pulses with a repetition rate of 100Hz are fed to the antenna input. The upper track is in logaritmic scale and the lower is in linear volts scale, red and green for I and Q respectively. This curve is essentially determined by the anti-alias filter.

Fig 5. Pulse response of direct conversion receiver tuned at 7MHz. Short pulses with a repetition rate of 100Hz are fed to the antenna input. The upper track is in logaritmic scale and the lower is in linear volts scale magnified 256 times (47dB). Red and green for I and Q respectively

Calibration procedure

The Linux DSP radio has to be calibrated by use of a pulse generator. The routines to collect the pulses are already in place (Aug 12 2000) and the next step is to include routines that collect the average shape of the fourier transforms of the pulses. The average pulse transform can then be inverted and multiplied by an ideal filter transform to produce the desired correcting filter.

Before starting to add more software I have to build a new pulse generator with lower repetition rate. Such a pulse generator will be a nessecary calibration tool for the Linux DSP radio.